Legal research.
With the source in view.
A legal research platform for a large law firm. Cursopic connected a 130 TB document archive to a workspace for cited answers, source verification, document review and drafting.
See the workspaceA vast archive.
A slow search.
The firm's archive held millions of decisions, claims and client files. Lawyers searched folders and databases, then cross-checked the source before using it in a case.
The task was to make that knowledge useful in daily work: searchable scans, checkable answers and access scoped to each matter.
Partners needed precedent for case strategy. Associates needed research and first drafts; paralegals needed review and summaries. The workspace had to connect those jobs to the same archive.
One firm's source archive.
Millions of documents.
Digital files and scanned pages.
What Cursopic built
Ingestion and OCRFrom source documents to searchable, citable passages.
Matter-level accessPermission-aware retrieval across the archive.
The lawyers' workspaceResearch, source verification, comparison and drafting.
Ask a question.
Follow the source.
A lawyer asks in everyday language. The workspace returns an answer with paragraph citations, then opens the original source for inspection. Scans follow the same path because OCR happens during ingestion.
Search the matters the lawyer can access. If no supporting passage is found, the system withholds an answer.
A whole system.
Behind each answer.
We connected the archive, access rules and the language model. Each step preserves the route from a question back to the evidence behind it.
Ingest and index
Scans go through OCR. Documents become paragraph-level passages in a vector index, with their metadata and matter permissions attached.
Retrieve permitted sources
A question is searched by meaning within the matters the user can access. The permission filter runs before passages are retrieved for the model.
Answer with citations
A cloud language model composes an answer from the retrieved passages. Citations open the original paragraph; the lawyer assesses the source and the conclusion.
The decisions that
make it useful.
The model was one part of the build. Source verification, matter boundaries and the daily workflow shaped the product around it.
Sources you can follow.
A citation opens the original paragraph, with document details alongside.
Access before answers.
Matter permissions are applied before passages are retrieved.
Engineering where it matters.
Custom engineering connects ingestion, access rules and sources across the workspace.
From research
to the next draft.
Two versions can be reviewed side by side, and a summary keeps its route back to the source. A new claim or letter starts from the firm's precedent, with the lawyer deciding which suggested passages stay.
Read two versions side by side, with changed clauses marked and a summary linked to the source.
Built. Launched.
Supported in daily use.
Our team took the platform from archive ingestion to production in two months, then stayed for a month of support as the firm moved its research into the workspace.
- 2months of development
From archive to workspace.
Ingestion and OCR made the documents searchable. Matter permissions and paragraph references carried through the index into research, review and drafting.
The archive layer and the lawyers' workspace were delivered as one working system.
- 1month of support
Stay with the daily work.
After launch, we watched how the firm used the platform, fixed issues reported by lawyers and tuned the results on their actual questions.
Feedback from real legal work came back into the product during the support month.
Less searching.
More time for the case.
Results reported in the project account.
More cases closed per lawyer
The firm's own estimate of output per lawyer.
Less time drafting documents
Reported average reduction in drafting time.
People using it
Lawyers and staff across the firm.
Results reflect the firm's project account. At that time, no production incidents had been reported since launch.
